138 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO.

1, JANUARY 2018

A Fully Parallel Stochastic Multiarea Power System

Operation Considering Large-Scale
Wind Power Integration
Mojtaba Khanabadi, Student Member, IEEE, Yong Fu , Senior Member, IEEE, and Lin Gong, Student Member, IEEE

Abstract—Multiarea power system operation/coordination is power systems can improve the economic aspects in operation
used to increase the reliability of the interconnected power grids of the entire system [2], [3], such as 1) lower overall congestion
and maintain the consistency of the price across the integrated costs; 2) more consistent prices across the areas; and 3) lower
power systems. However, the implementation of this coordinated
operation is facing challenges due to increase in size and com- operating cost due to presence of broader pool of mutual benefits
plexity of the modern power systems, high penetration of the shared between systems.
volatile renewable energy, and interdependence issues among vari- A traditional approach to providing such a coordination be-
ous power systems. To address these concerns, this paper presents a tween the interconnected power systems is through forming a
fully parallel decision-making approach for the day-ahead schedul- power pool [1]. In this method, a third party utilizes the de-
ing of interconnected power systems with large-scale wind power
integration while respecting the information privacy between tailed information submitted by different power systems, such
different systems/areas. In the proposed parallel approach, each as characteristics of generating units, availability and parameters
system/area solves its day-ahead scheduling problem along with its of transmission networks, predicted demand profiles, and output
local subproblems for different wind generations’ scenarios, and power profile of wind turbines across the interconnected power
sends its equivalent (or processed) boundary information to other systems [4], to formulate and solve a large-scale generation
systems/areas. The proposed inter-regional coordination among
systems/areas and intra-regional coordination between scenarios scheduling problem over the entire interconnected power sys-
in each system/area will continue until the tie-line power flows and tems. However, such a solution approach is centralized, mostly
the generation outputs of generating units get converged. The mod- very complicated and time-consuming due to enormous num-
ified IEEE 118-bus testing system is used in this paper to show the bers of decision variables and operational constraints of the
effectiveness of the proposed approach. problem. In the advanced operation of interconnected power
Index Terms—Parallel optimization, multiarea coordination, systems, the system operators need to consider transmission
security-constrained unit commitment, wind power integration. constraints into their regional operation which requires the inte-
grated systems to exchange their grids’ information like unit’s
I. INTRODUCTION generations, network topology, and loads [3]. However, system
operators are usually reluctant to share their system information
OWADAYS, power systems are interconnected together
N to provide a reliable and secure power supply to customers
and operate in lower operating cost than if they are being run
as this information is counted as private and might be commer-
cially sensitive and confidential to other areas. Therefore, one
of the motivations of this paper is to provide a methodology
separately [1]. In the event of an emergency in any individ- which enables different system operators to efficiently sched-
ual system, such as a shortage of generation capacity and/or ule their regional generation resources and optimally coordinate
an unexpected change in wind power generation, the intercon- their operations with other neighboring areas without suffering
nected power system can utilize all available power generation from the shortcomings of the centralized solutions and infor-
resources and delivery facilities throughout the entire grid to mation exchanging. Our proposed idea relies on decentralized
adjust the transferring power among systems, thus guaranteeing solution methodologies. Using such methods, the original large-
a continuous power supply to customers and achieving a high- scale problem can be divided into several scalable and tractable
level power system reliability. In addition, interconnecting the area-based subproblems that can be coordinated with each other
to find out the optimal operating condition of the entire power
Manuscript received November 29, 2016; revised April 16, 2017; accepted system while respecting information privacy [5].
June 4, 2017. Date of publication June 26, 2017; date of current version Decem-
ber 14, 2017. This work was supported by the U.S. National Science Foundation Additionally, in the current operation of power systems, wind
under Grant ECCS-1150555. Paper no. TSTE-00939-2016. (Corresponding au- power is one of the most favorable sources of energy to serve the
thor: Yong Fu.) costumers. It has been evident that the large-scale integration of
The authors are with the Department of Electrical and Computer Engineer-
ing, Mississippi State University, Mississippi State, MS 39762 USA (e-mail: the wind power would dramatically contribute to power system
[email protected]; [email protected]; [email protected]). economics [6]. However, high penetration of wind power raises
Color versions of one or more of the figures in this paper are available online serious concerns during the operation of power systems. Most
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSTE.2017.2719659 of these concerns, in one way or another, are related to uncertain

1949-3029 © 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
KHANABADI et al.: FULLY PARALLEL STOCHASTIC MULTIAREA POWER SYSTEM OPERATION 139

nature of the wind. In the case of unexpected changes in wind two neighboring areas and forced the duplicated variables (e.g.,
power generation, power system operators would like to be able bus voltage angles) for that fictitious bus to have an almost
to mitigate the violations accordingly and provide continuous same value at the final solution point of the OPF problem. As an
power supply to the customers without any interruption in ser- extension of [22], an auxiliary problem principle (APP) based
vice. To solve such uncertainty issue, different methods have optimization approach was presented in [23] to enhance the
been proposed by researchers to effectively reduce the com- convergence performance and efficiency of the decentralized
plexity of the stochastic problems. In [7] and [8], a chanced solution process of OPF problem. In contrast to [22] and [23]
constrained optimization was proposed to deal with intermit- where the decoupling is executing around fictitious buses in
tent nature of wind power generation where the system security boundary areas, [24] decouples the DC-OPF problem around tie
constraints can be violated with a small probability index. Ro- lines. Using this approach, each area just needs to exchange tie
bust optimization has also widely utilized to model the wind lines information with its neighboring areas. An analytical tar-
power generation uncertainties by providing a span around the get cascading (ATC) based decentralized solution methodology
central forecast [9], [10]. The stochastic programming is also for the large-scale SCUC problem was presented in [25] to find
one of the most promising optimization tools available to sys- a feasible and optimal operating point across the entire power
tem operators, in which possible realizations of uncertainties system. In [25], each regional system should iteratively send its
can be simulated in their scheduling problems. [11] proposed boundary bus voltage angles and tie-line power flows to a higher
a stochastic based unit commitment scheme to examine the level coordinator in order to coordinate its operation with other
impact of wind power generation and load variations on dis- neighboring regional systems. [26] presented a decentralized
patching result of a power system. [12] presented a two-state methodology to optimally schedule generating units in a multi-
stochastic programming to guarantee enough reserve allocation area market-clearing operation while considering uncertainties.
in a power system with large amount of wind power genera- To deal with the uncertainty nature of wind energy generation,
tion. [13] studied a stochastic operation planning of the power [27] proposed a decentralized methodology for day-ahead en-
system based on the market-clearing with stochastic security in ergy and reserve market clearing in a multi-area power system by
the presence of wind and demand uncertainty. [14] improved introducing the adjustable interval robust scheduling based on
the stochastic programming approach in the unit commitment the interval optimization approach. Authors in [28] proposed an
problem to incorporate wind power scenarios by introducing a APP based Lagrangian relaxation method to simultaneously co-
dynamic decision making approach. Note that the quality of re- ordinate the day-ahead scheduling of both the inland HVAC and
sult in the stochastic optimization relies heavily on the number offshore HVDC power systems, while considering the offshore
of scenarios considered in the problem formulation. Normally, wind uncertainties using stochastic programming technique.
a higher number of the scenarios would lead to a more accurate In this paper, a new decentralized multi-area coordination for
result and also increase in size and complexity of the problem the stochastic SCUC of interconnected power systems is pro-
formulation [15], [16]. As a result, the computational burden posed. The contributions of this paper in terms of tie-line power
of the stochastic optimization problem can be increased until flow modeling and inter/intra-regional decomposition and
the problem becomes huge and unsolvable. Avoiding this issue, coordination methods are summarized as follows:
different scenario reduction techniques have been employed by 1) Conventionally, the difference between voltage angles at
literature to effectively reduce the number of scenarios while both ending buses of the tie line is used to determine the
ensuring high quality of the results [17]–[21] (e.g., the K-means power flow exchanges between areas, and thus the volt-
clustering algorithm [17], the submodular function optimization age angles at the boundary buses were commonly used to
method [18], the functional approximation in Kantorovich dis- model the shared variables/exchanged data among areas
tance [19], the scenario tree construction algorithm [20], and the [22]–[27]. Such a tie-line power flow modeling requires a
improved forward-selection and backward-reduction approach decentralized SCUC solution tool to completely model all
[21]). However, the stochastic optimization with reduced num- bus voltage angles and transmission lines across entire in-
ber of scenarios can be still complicated enough to demand terconnected power systems, which may increase the size
considerable computation efforts. Therefore, another motiva- of the problem and the required solution time. However,
tion of this paper is to propose a method to overwhelm this in this paper, we utilize a shift factor (SF) based power
shortcoming. To do so, in this paper, the decentralized solution flow representation to model the power transfers through
methodologies are further utilized to decompose the regional tie line and coordinate exchanging power between areas.
stochastic day-ahead scheduling problem of systems/areas into This proposed alternative tie-line power flow model in our
multiple subproblems (one for each scenario). Using this tech- decentralized SCUC solution framework makes it possible
nique, each scenario can be formulated and solved in parallel to only formulate and monitor just certain critical trans-
with its corresponding deterministic subproblem; therefore, the mission lines, especially for a large-scale power system,
overall solution procedure can be accelerated significantly. thus, which could result in reducing the size and acceler-
The decentralized multi-systems/areas coordination has been ating the solution procedure.
previously investigated in the large-scale power system schedul- 2) In addition, despite existing publications [11]–[14] in
ing problems, such as optimal power flow (OPF) [22]–[24] and which the stochastic day-ahead scheduling of a single
security-constrained unit commitment (SCUC) [25]–[28]. [22] area power system was discussed or publications [26] and
introduced a fictitious bus inside an overlapping zone between [27] in which the wind power generation scenarios were
140 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 1, JANUARY 2018

processed in a centralized manner in operation of individ-

ual areas, in this paper, both inter/intra-regional decompo-
sition and coordination strategies are proposed to divide
the original stochastic multi-area SCUC (SMA-SCUC)
into multiple subproblems in terms of both areas and sce-
narios. Here, the individual areas’ scheduling subproblem
will be formulated and solved in parallel. Then, the re-
gional stochastic scheduling subproblems can be further
divided into a base case subproblem and multiple subprob-
lems for different scenarios, and those subproblems can be
also solved in parallel to further accelerate the proposed
Fig. 1. An interconnected power system with multiple individual areas.
solution procedure. As a result, lower execution time and
higher solution procedure’s efficiency can be achieved.
The rest of the paper is organized as follows: Section II in- cost of the interconnected power systems over the studied time
troduces a generic SMA-SCUC formula. Section III presents horizon (e.g., 24 hours in this paper) for all scenarios. The
inter/intra-regional decomposition and coordination strategies, constraints (1b) are relevant to minimum On/Off time limits and
and discusses the parallel solution procedure. Section IV uses ramping up/down limits of generating units. The constraints (1c)
the modified IEEE 118-bus testing system to evaluate the ef- represent the generation capacity limits of generating units. The
fectiveness of the proposed method. A conclusion is drawn in constraints (1d) include the power balance of the entire system
Section V. (2a) and the shift factor based power flow limits of transmission
lines (2b) within all the interconnected power systems.
II. GENERIC SMA-SCUC PROBLEM MODELING NG
 N
 W ND

As one of the most important power system operation Pgs + Pws − Di = 0 ∀s (2a)
decision-making tools, SCUC is commonly being referred as g=1 w =1 i=1

scheduling of energy resources to satisfy demands in the most |PLs = GSF(Ps − D)| ≤ PLm ax ∀s (2b)
economical way while meeting the system’s security criteria
[29]. In this section, as shown in Fig. 1, an interconnected power where, Pgs is the generation output of conventional thermal gen-
system with N individual areas is used to introduce the proposed erating unit g, Pws is the generation output of wind generating
SMA-SCUC model considering multiple wind generation sce- unit w, and Di is the expected load consumption of load i; the
narios. As illustrated in Fig. 1, any of Areas 1 to N can be total number of conventional thermal generating units, wind
connected to its neighboring areas through a tie line (could be generating units, and loads in the entire interconnected power
multiple tie lines). system are noted by NG, NW and ND, respectively; Ps and D
Mathematically, the proposed SMA-SCUC can be formulated are the generations (of both conventional thermal and wind gen-
as a mixed integer programming (MIP) problem. Without loss erating units) and loads within the entire system, respectively;
of generality, a set of general SMA-SCUC formula (1) can be PLs and PLm ax represent the line power flows and their capac-
expressed as follows (note that the notation for time t is omitted ity limits, respectively; and GSF is defined as the global shift
in order to simplify the model’s expression and focus on the factors for the transmission lines.
proposed decomposition and coordination methodologies in the In the above SMA-SCUC problem, the objective function (1a)
paper), has no coupling terms between areas/scenarios; thus, it is de-
composable. However, the constraints (1b)–(1d) are making the
N 
 NS
entire optimization problem (1) complicated and indecompos-
Min ωξs Fξ (Iξ , Pξs ) (1a)
able in terms of areas and scenarios. For example, the constraints
ξ =1 s = 0
(1d) are relevant to all areas, and all of the scenarios must have
S.t. Asξ Iξ + Bsξ Psξ ≤ dsξ ∀ξ, ∀s (1b) the same generating units’ status I. The details of our proposed
decomposition and coordination strategies are discussed in the
Pm in,ξ · Iξ ≤ Psξ ≤ Pm ax,ξ · Iξ ∀ξ, ∀s (1c) following sections.
N

Esξ Psξ ≤ hs ∀s (1d) III. INTER/INTRA-REGIONAL DECOMPOSITION AND
ξ =1 COORDINATION STRATEGIES
where, s indicates the scenario (s = 0 for the base case, and the This section will introduce inter/intra-regional decomposition
total number of scenarios is NS); ξ is the index of areas; ωξs is and coordination strategies to implement the proposed SMA-

the probability of scenarios and is subject to ωξ0 + s= 0 ωξs = SCUC in a parallel manner. As illustrated in Fig. 2, at the first
1;Iξ is the commitment of generating units (On/Off) in Area inter-regional decomposition and coordination stage, an equiva-
ξ for all scenarios; and Psξ is the active power generation of lent (or a processed) boundary information is proposed to create
generating units in Area ξ for scenario s. The objective (1a) of the interactions among areas. Next, in order to accelerate the
the proposed SMA-SCUC is to minimize the expected operating proposed solution procedure, an intra-regional decomposition
KHANABADI et al.: FULLY PARALLEL STOCHASTIC MULTIAREA POWER SYSTEM OPERATION 141

system (e.g., KCL and KVL). This fact is expressed in (5).

PLstie J = GSFtie J (Ps − D)
⎡ ⎤
Ps1 − D1
⎢ .. ⎥
⎢ ⎥
 ⎢ s . ⎥
= GSFtie J ,1 . . . GSFtie J ,J . . . GSFtie J ,N ⎢ PJ − DJ ⎥
⎢
⎥
⎢ .. ⎥
⎣ . ⎦
PsN − DN
= GSFtie J ,1 (Ps1 −D1 )+ · · ·+ GSFtie J ,J (PsJ − DJ )+ · · ·
+ GSFtie J ,N (PsN − DN ) ∀s
(5)
where GSFtie J is the global shift factor for the tie lines con-
nected to Area J, and its submatrices GSFtie J ,1 , GSFtie J ,J ,
to GSFtie J ,N are associated with Areas 1, J, to N, respectively;
Fig. 2. Illustration of inter/intra-regional decomposition and coordination Ps1 , PsJ , to PsN and D1 , DJ , and DN are the generations and
strategies.
loads associated with their corresponding areas.
Realize from the problem modeling presented above, the only
and coordination strategy is proposed to further reformulate the coupling constraint between areas is the tie-line power flow (5),
relationship between the base case and scenarios. which consists of N components (one from each area). These
components can be regarded as the contributions of individual
A. Modeling of Coupling Constraints for Inter-Regional areas to the tie-lines power flow. Therefore, from the viewpoint
Decomposition and Coordination Strategy of Area J, the power flow in tie-line tie which connects Area J
First of all, in order to successfully decompose the original to its neighboring area, and its corresponding components can
SMA-SCUC problem (1) into multiple scalable and tractable be expressed as follows:
regional stochastic SCUC subproblems, one for each area, N
 s
the key thing is how to model the tie-line power flow. To pro- P Lstie J = P Lstie J ,J + P Ltie J ,ξ ∀s (6)
vide a generic model for each individual area, here, we focus ξ = J
on modeling of the tie lines just connected to Area J (it can be where P Lstie J ,J is the contribution of Area J to the power flow
easily applied for any other individual area). For Area J, the s
of tie-line tie; P Ltie J ,ξ is regarded as the expected/estimated
power balance constraint (2a) can be rewritten as follows,
contribution of other Area ξ to the tie-line power flow. Note
N
 GJ N
WJ N
 DJ N
 BJ  that, due to information privacy between areas, Area J does not
Pgs + Pws − Di = Kb,k · P Lsk ∀s have the access to local generations and consumptions in other
g=1 w =1 i=1 b = 1 k ∈tie J areas. Accordingly, Area J cannot directly calculate and include
(3) other areas’ contributions to its tie-line power flow calculation
where the total number of conventional thermal generating units, in (6). To handle this issue, Area J is required to estimate other
wind generating units, loads, and buses in Area J are noted by areas’ contributions to its connected tie-lines. Therefore, these
N GJ , N WJ , N DJ and N BJ , respectively; P Lsk is the power expected/estimated contributions are indicated by the overline
flow of tie-line k transferring power from Area J to its neighbor- “-” in our model.
ing areas during scenario s; K is the tie line k to regional bus b As the physical law of the power systems (e.g., KVL and
incident. And, the shift factor based power flow limits (2b) for KCL) indicates that the contribution of each area to the tie-line
regional transmission lines inside Area J can be reformulated as power flow should have the same value as the contribution of
follows, this area which is expected/estimated by other areas. Therefore,
 s 
PLJ = SFJ (PsJ − DJ − Ktie · PLstie ) ≤ PLm ax,J ∀s the following equalities should be met in the proposed model
J
s
(4) P Lstie J ,J = P Ltie J ,J ∀s (7a)
where PLsJ is the power flows through regional transmission
s
lines of Area J; PsJ is the generations associated with Area J; DJ P Ltie J ,ξ = P Lstie J ,ξ ∀s, ξ = J (7b)
is the expected loads within Area J; PLstie J is the power flows
Note that in (7), the tie-line power flows’ components without
in tie-lines connected to Area J; SFJ is the regional/local shift
the overline “-” can be exactly determined by using regional
factor matrix associated with Area J; PLm ax,J is the maximum
areas’ information (e.g., regional generations and loads) (8):
power flow through Area J’s regional transmission lines.
However, just to model the tie-line power flow as a pseudo P Lstie J ,J = GSFtie J ,J (PsJ − DJ ) (8a)
load (or power withdrawal) is not sufficient because the tie-line
power flow PLstie J must follow the physical law of the power P Lstie J ,ξ = GSFtie J ,ξ (Psξ − Dξ ) (8b)
142 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 1, JANUARY 2018

Therefore, once all the above equalities (7) and (8) are for base case constraints (10b-c) and the other one for scenar-
satisfied, we can guarantee that the tie-line power flows can ios constraints (10d-e). However, since the vector of generating
satisfy the physical law (e.g., KCL and KVL) of the entire in- units’ status in base case constraints (I0J ) and scenarios con-
terconnected power system. Now, the coupling constraint (5) straints (IsJ ) are not coupled to each other, one can find that
is replaced by the coupling constraint (7) which create the the status of generating units in base case and scenarios sub-
interactions between regional areas. problems may not be equal to each other. Avoiding this issue,
we have introduced a new set of continuous variables (P0,s J )
B. Modeling of Coupling Constraints for Intra-Regional for each scenario, which can be regarded as the duplication of
Decomposition and Coordination Strategy the generating units’ outputs in the base case (P0J ), and a set
of constraints shown by (10f) in the problem. Mathematically
Notice that the power flows in the tie lines connected to Area
speaking, once these duplicated variables become equal to their
J is represented by the variable PLstie J . Therefore, Area J’s
corresponding values in the base case subproblem (the equality
stochastic SCUC problem can be generalized as follows:
(10g) holds), the vectors of generating units’ statuses (I0J and
NS
 IsJ ) become accordingly equal to each other. Now, the coupling
Min ωJs FJ (IJ , PJs ) (9a) constraint (10g) is formulated to create the relationship between
s=0 the base case and scenarios in Area J.
S.t. HsJ IJ + LsJ PsJ + MsJ Pstie J ≤ esJ ∀s (9b)
Pm in,J · IJ ≤ PsJ ≤ Pm ax,J · IJ ∀s (9c) C. Regional SCUC Formula
In the previous Sections III.A and III.B, we have intro-
In (9), the objective function (9a) is to minimize the ex-
duced two different sets of coupling constraints: one (7) for
pected operating cost of Area J over the studied time horizon;
inter-regional decomposition in which we divided a large-scale
the linear constraint (9b) is the set of Area J’s equality and in-
stochastic SCUC into multiple regional stochastic SCUC sub-
equality constraints, such as minimum On/Off time limits and
problems, and the other (10g) for intra-regional decomposition
ramping Up/Down limits of the local generating units, regional
in which each regional stochastic SCUC subproblem is decom-
power balance (3) and power flows limits (4), and the tie-line
posed into multiple smaller subproblems (one for each scenario).
power flow limits relevant to Area J; the constraint (9c) is for
Here, the augmented Lagrangian relaxation method is employed
the generation capacity of generating units within Area J. As
to relax the coupling constraints (7) and (10g) by adding them as
we discussed earlier, including the wind generation scenarios
a first-order and second-order penalty functions into the objec-
would lead to a significant increase in size and complexity of
tive function (1a). Further, auxiliary problem principle (APP) [5]
the scheduling problems. In order to effectively decompose the
is adopted to handle the coupling terms (second-order penalty
regional stochastic SCUC subproblems into a base case sub-
functions) in the obtained Lagrangian objective function. Con-
problem and multiple scenarios subproblems, we reformulate
sequently, the regional base case problem of Area J is obtained
the problem (9) as follows,
as follows,
NS

Min ωJ0 FJ (IJ0 , PJ0 ) + ωJs FJ (IJs , PJs ) (10a) Min ωJ0 FJ (IJ0 , PJ0 )
s=1  2
+ ρ · P L0k ,J
S.t. k ∈tie J
 0,n −1

For base case subproblem: + αks ,J − ρ P L0,n
k ,J
−1
+ P Lk ,J P L0k ,J
 
H0J I0J + L0J P0J + M0J P0tie J ≤ e0J (10b) + ρ · P L0k ,ξ
2

Pm in,J · I0J ≤ P0J ≤ Pm ax,J · I0J (10c) ξ = J k ∈tie J

 0,n −1  0
For scenarios subproblems: + αks ,ξ − ρ P Lk ,ξ + P L0,n
k ,ξ
−1
P Lk ,ξ
 2
HsJ IsJ + LsJ PsJ + MsJ Pstie J ≤ esJ ∀s = 0 (10d) + ρ · P L0k ,J
k ∈tie, k ∈
/ tie J
Pm in,J · IsJ ≤ PsJ ≤ Pm ax,J · IsJ ∀s = 0 (10e)  
0,n −1
+ αks − ρ P L0,n −1
+ P Lk f r om ,J P L0k ,J
Pm in,J · IsJ ≤ P0,s
J ≤ Pm ax,J · IsJ ∀s = 0 (10f) f r om , J k ,J
 2
Coupling constraints: + ρ · P L0k J
k ∈tie, k ∈
/ tie J
P0J = P0,s
J ∀s = 0 (10g)  0,n −1

+ αks to, J − ρ P L0,n
k ,J
−1
+ P Lk to,J P L0k ,J
As evidenced by (10a), the objective function of each individ-
ual area’s stochastic SCUC can be divided into a base case term  N
GJ
2   
along with multiple scenarios terms. Also, in order to make the + ρ·Pg0 + βgs − ρ Pg0,n −1 + Pg0,s,n −1 Pg0
problem (9) decomposable, the prevailing constraints (9b) and s= 0 g = 1

(9c) can be further divided into two groups of constraints, one (11)
KHANABADI et al.: FULLY PARALLEL STOCHASTIC MULTIAREA POWER SYSTEM OPERATION 143

Fig. 3. Decomposition structure of the proposed fully parallel solution.

Subject to (10b), (10c), (6) and (8a) for base case (s = 0).
In addition, the regional subproblems of Area J associated
with scenario s is written as follows:
Min ωJs FJ (IJs , PJs )

+ ρ · P Lsk ,J 2
k ∈tie J
 s,n −1

+ αks ,J − ρ P Ls,n
k ,J
−1
+ P Lk ,J P Lsk ,J
 
+ ρ · P Lsk ,ξ 2
ξ = J k ∈tie J
 s,n −1  s
+ αks ,ξ − ρ P Lk ,ξ + P Ls,n
k ,ξ
−1
P Lk ,ξ

+ ρ · P Lsk ,J 2
k ∈tie, k ∈
/ tie J
Fig. 4. Flowchart of the proposed parallel approach.
 s,n −1

+ αks f r om , J −ρ P Ls,n
k ,J
−1
+ P Lk f r om ,J P Lsk ,J
D. Stochastic Multi-Area SCUC Solution Procedure

+ ρ· P Lsk ,J 2 The major solution steps for the implementation of the
k ∈tie, k ∈
/ tie J proposed inter/intra-regional decomposition and coordination
 s,n −1
 strategies are shown in Fig. 4.
+ αks to, J − ρ P Ls,n
k ,J
−1
+ P Lk to,J P Lsk ,J These steps are summarized as follows:
N GJ
Step 1: Set the iteration index n = 0 and choose initial
 2   
+ ρ · P 0,s + βg0 − ρ Pg0,n −1 + Pg0,s,n −1 Pg0,s values for all exchanged information (e.g., tie-lines
g
g=1
power flows’ components and generating units’ out-
(12) puts).
Step 2: Set n = n + 1, solve the regional subproblems (11)
Subject to (10d), (10e), (10f), (6) and (8a) for scenario s and (12) all in parallel using the updated information
(sࣔ0). from other areas and/or base case/scenarios subprob-
Note that the variables P Lk f r om ,J and P Lk to,J in (11) lems.
and (12) are the estimated contribution of Area J to the power Step 3: Check the following stopping criteria:
flow of tie line k that is not connected to Area J, from the Stopping criterion 1: Each area’s contribution to the tie-line
viewpoint of the tie line k’s sending-end and receiving-end areas, power flow (P Ltie,ξ ) should have almost same value as its
respectively. In both (11) and (12), parameters α and β are the contribution which is expected by its other areas (P Ltie,ξ ).
first order penalty functions’ multipliers; and the parameter ρ
is the positive second-order penalty functions’ multipliers. The  
 s,n s,n 
superscript n-1 in (11) and (12) is used to indicate the values P Ltie,ξ − P Ltie,ξ  ≤ ε1 ∀s, ∀ξ (13)
which are obtained from the previous iteration.
Similarly, the regional SCUC base case and scenarios prob- Stopping criterion 2: The outputs of generating units in the
lem formula for all other areas can be formulated without any base case subproblems should be almost equal to their duplica-
difficulties. Finally, as shown in Fig. 3 the original multi-area tions in the scenario subproblems.
stochastic SCUC problem is decomposed into multiple scalable  0,n 
and tractable subproblems in terms of both areas and scenarios. Pg − Pg0,s,n  ≤ ε2 ∀s, ∀g ∈ N Gξ , ∀ξ (14)
144 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 1, JANUARY 2018

been widely utilized by researchers to cluster a huge amount of

physical processes data into fewer numbers due to its simplicity
and good performance [31], [32]. However, any other scenario
generation/reduction methods with acceptable accuracy could
be used here to generate/reduce corresponding scenarios which
can be used as inputs to our work. The following three cases are
discussed:
Case 1: Deterministic Multi-Area Coordination: In this
case, the expected outputs of wind generating units are used
for the base case. Since no wind scenario is considered in the
problem, three areas are required to coordinate their operating
condition together just for the deterministic condition. Fig. 6
depicted the trend of the change in the total power mismatch
(absolute value) of (13) across the entire interconnected power
system. As we can see, the proposed parallel method converges
and all stopping criteria (13) are met after 78 iterations. The
Fig. 5. One-line diagram of the IEEE 118-bus testing system with three obtained total operating cost of interconnected power systems
individual areas.
is $1,584,426 which is just 0.25% higher than the operating cost
of the conventional centralized solution.
If the above stopping criteria are met, then stop the solution Here, in order to further discuss the convergence performance
procedure; otherwise, update the penalty multipliers using (15) of the proposed decentralized method, a comparison between the
and go to Step 2. actual and expected/estimated components (or contributions) of
 s,n
 the power flow in the tie-line 70-69 (connecting Area 1 and
s,n +1 s,n
αtie,ξ = αtie,ξ + ρn P Ls,n
tie,ξ − P L tie,ξ ∀s, ∀ξ (15a)
Area 2) at Hour 18 is provided in Figs. 7–9. According to
  (6), for an interconnected power system with three areas, the
βgs,n +1 = βgs,n + ρn Pg0,n − Pg0,s,n ∀s, ∀g ∈ N Gξ , ∀ξ
tie-line’s power flow consists of three components. Therefore,
(15b)
from the viewpoint of Area 1, the power flow of tie-line 70-69
ρn + 1 = ρn · γ (15c) (P L70−69 1 ) and its components can be shown as follows:
where γ ≥ 1. P L70 − 69 1 = P L70−69,1 + P L70−69,2 + P L70−69,3
where P L70−69,1 is the contribution of Area 1 to the power
IV. CASE STUDIES flow of tie-line 70-69, P L70−69,2 and P L70−69,3 are the ex-
In this section, the modified IEEE 118-bus testing system is pected/estimated contributions of Area 2 and Area 3 to the power
employed to illustrate the effectiveness of the proposed paral- flow of tie-line 70-69 from the viewpoint of Area 1.
lel stochastic multi-area SCUC approach. For all case studies, Mathematically speaking, when a converged result is ob-
the initial values of exchanged information in (11) and (12) tained, these three components (or contributions) should be
are set to 0 (flat start), and the thresholds for stopping criteria equal to their corresponding expected/actual contributions cal-
(13) and (14) are ε1 = ε2 = 0.1 M W . Also, ρ and γ are set to culated by other neighboring areas: P L70−69,1 = P L70−69,1 ,
0.05 and 1.01, respectively. All simulations are conducted on a P L70−69,2 = P L70−69,2 , and P L70−69,3 = P L70−69,3 , where
2.8 GHz personal computer, and the ILOG CPLEX 12.6.3 solver P L70−69,1 is the expected contribution of Area 1 to the power
is employed to solve the resulted optimization problems. flow of tie-line 70-69 from the viewpoint of Area 2; P L70−69,2
The one-line diagram of the studied IEEE 118-bus testing and P L70−69,3 are the actual contributions of Area 2 and Area
system is shown in Fig. 5 where the system compasses three in- 3 to the power flow of tie-line 70-69. As the simulation results
dividual areas. More information regarding the areas’ properties shown in Figs. 7–9, as the iteration number of the proposed de-
are given in Table I. In this system, Area 1 is connected to Area centralized method increases, the gap between above actual and
2 through 7 tie lines; Area 2 is connected to Area 3 through 5 tie expected/estimated values decreases. Finally, at iteration 78, all
lines, and no tie lines between Area 1 and Area 3. In this study, the stopping criteria will be met.
the IEEE 118-bus testing system encompasses 10 wind generat- Case 2: System Operation Under Wind Uncertainty Without
ing units. Here, to achieve a more accurate random distribution, Multi-Area Coordination: In this case, let us assume that the
the Monte Carlo Simulation (MCS) along with Latin hyper- generating units’ status obtained from Case 1 (deterministic)
cube sampling technique is employed to generate 1,000 wind is used; and also the power flows through tie-lines are fixed to
power generation scenarios [30]. These scenarios will represent their values calculated by Case 1. As a result, each individual
possible realizations of the stochastic process during day-ahead area can be considered as an autonomous entity and is required
scheduling of interconnected power systems. Then, to reduce the to locally handle its’ local power mismatch resulted by the
number of generated scenarios to different numbers for testing deviation between the expected and real wind power gener-
(e.g., 1, 2, 5, 10 and 20 scenarios in case studies), the well- ation. As an example, during wind scenario No. 1, Areas 1
known K-means clustering algorithm is employed since it has and 3 can successfully balance their power mismatches locally.
KHANABADI et al.: FULLY PARALLEL STOCHASTIC MULTIAREA POWER SYSTEM OPERATION 145

TABLE I
INDIVIDUAL AREAS’ INFORMATION IN THE IEEE 118-BUS TESTING SYSTEM

Areas # of Buses # of Thermal Units # of Wind Units # of Branches Max Gen. (MW) Peak Demand (MW) # of Tie Lines between Areas

Area 1 Area 2 Area 3

Area 1 42 19 4 57 46,800 33,488.5 N/A 7 0

Area 2 48 20 4 75 83,520 54,639.5 7 N/A 5
Area 3 28 15 2 43 42,960 22,024.4 0 5 N/A

Fig. 6. Total power mismatch of (13) during the iterative solution procedure. Fig. 9. Comparison between the contribution of Area 3 to the power flow
of tie-line 70-69 which is expected/estimated by Area 1 (P L 7 0 −6 9 , 3 ) and the
contribution of Area 3 from the viewpoint of Area 3 (P L 7 0 −6 9 , 3 ).

Fig. 7. Comparison between the actual contribution of Area 1 to the power

flow of tie-line 70-69 from the viewpoint of Area 1 (P L 7 0 −6 9 , 1 ) and the Fig. 10. Total generations (G), demands (D), and load shedding amounts (LS)
in each individual area under wind scenario (s = 1) at Hour 13.
contribution of Area 1 expected/estimated by Area 2 (P L 7 0 −6 9 , 1 ).
TABLE II
STOCHASTIC MULTI-AREA SCUC RESULTS

# of Decentralized Centralized Gap (%)

scenarios

Total Cost Exe. time # of Total Cost Exe. time

($) (sec.) iterations ($) (sec.)

0 1,584,426 45.6 78 1,579,826 4.5 0.30

1 1,585,130 71.4 69 1,580,390 5.11 0.29
2 1,585,189 89.1 90 1,581,699 48.20 0.22
5 1,592,444 151.2 127 1,589,967 583.61 0.15
10 1,600,276 344.9 241 N/A N/A N/A
20 1,621,020 562.5 244 N/A N/A N/A

Fig. 8. Comparison between the contribution of Area 2 to the power flow

of tie-line 70-69 which is expected/estimated by Area 1 (P L 7 0 −6 9 , 2 ) and the
actual contribution of Area 2 from the viewpoint of Area 2 (P L 7 0 −6 9 , 2 ).
Case 3: With Inter/Intra-Regional Coordination: for this
case study, three areas can coordinate their operating points
However, Area 2 has to perform load shedding of 2.42 WM and with each other for all scenarios. Table II lists the results of the
34.95 MW at Hours 9 and 13, respectively. Fig. 10 depicts the proposed parallel approach for a different number of the wind
total generations, demands, and load shedding amounts across generation scenarios, including the total operating cost of the in-
the interconnected power systems at Hour 13. terconnected power systems, the required number of iterations,
146 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 9, NO. 1, JANUARY 2018

Fig. 11. Converged percentage of the inter/intra-regional coupling constraints over the 118 bus interconnected power systems.

TABLE III Also, Table III shows the transferring power between individ-
INDIVIDUAL AREAS’ EXCHANGING POWER UNDER 20 SCENARIOS
ual areas under 20 different scenarios (plus base case 0) at Hour
18. As we can see, due to volatile nature of wind power genera-
Scenario Net Power Flow Between Scenario Net Power Flow Between tion, the transferring power between individual areas is different
No. Area 1 & 2 Area 2 & 3 No. Area 1 & 2 Area 2 & 3 for each scenario, which makes it possible for the system opera-
(MW) (MW) (MW) (MW) tor to retain an economic and secure operating condition during
0 −244.5 −134.2 11 −241.7 −132.8
any of these scenarios for the interconnected power systems.
1 −243.5 −133.8 12 −243.6 −133.0
2 −244.2 −134.2 13 −244.3 −134.2
3 −242.6 −133.3 14 −244.1 −134.0
4 −240.3 −132.0 15 −246.6 −135.5
V. CONCLUSION
5 −240.5 −132.1 16 −245.6 −134.1 The system operators are responsible for a reliable and se-
6 −241.6 −132.0 17 −240.5 −132.7
7 −239.9 −131.4 18 −238.6 −131.1 cure operation of power systems. In this paper, a decentralized
8 −242.1 −134.8 19 −241.7 −132.8 solution method based stochastic multi-area scheduling prob-
9 −244.0 −135.9 20 −244.3 −134.2 lem is presented and discussed where the individual areas co-
10 −244.8 -134.1 - - -
operate with each other to find the optimal operating point in
an interconnected power system with large-scale wind power
integration. In the proposed approach, two different decom-
and the execution time. Table II shows that the total operating position strategies are presented to firstly divide the original
cost of the interconnected power systems, calculated using our large-scale SMA-SCUC problem into multiple local stochastic
proposed parallel approach, is close to its centralized ones while SCUC problems; and then divide the local stochastic SCUC
considering 1, 2, and 5 scenarios. Note that the conventional cen- problem into multiple subproblems, one for the base case and
tralized approach cannot handle the stochastic multi-area SCUC one for each scenario. Using this approach, the system oper-
problem with 10 and 20 scenarios due to the size and complexity ators will be able to solve their local optimization problems
of the problem. However, using our proposed parallel approach, all in parallel, thus, the overall solution procedure can be ac-
the converged results for these two cases can be obtained in 241 celerated. The numerical tests on the modified IEEE 118-bus
and 244 iterations, respectively. testing system justified the effectiveness of the proposed ap-
To further show the convergence performance of the proposed proach. The results verified that using the proposed inter/intra-
parallel approach, the percentage of the converged inter/intra- regional coordination strategies, the optimal operating condition
regional coupling constraints (13) and (14) for different numbers over the entire interconnected power system can be achieved
of scenarios is shown in Fig. 11. As we can see, with an increase while mitigating uncertainties due to the large-scale wind energy
in the number of scenarios, the number of coupling constraints integration.
introduced in the problem formulation increases significantly. The proposed modeling strategies and parallel solution frame-
As an example, the case with 20 scenarios has 203,904 coupling work could be also used to 1) conduct a rapid analysis on reli-
constraints. Thus, the solution procedure may need a higher ability and economics of large-scale power system considering
number of iterations to get the converged result. As Fig. 11 multiple scenarios within multiple areas over multiple periods;
illustrates, roughly after running 50 iterations, more than 95% 2) promote coordination and interaction between regional power
percent of the coupling constraints for all five studied cases are systems; 3) identify how power system operations are affected
converged. Then, extra iterations are required for these cases to by the integration of wind energy, and thus, 4) facilitate a higher
satisfy all inter/intra-regional coupling constraints. penetration of large-scale wind energy into power grids.
KHANABADI et al.: FULLY PARALLEL STOCHASTIC MULTIAREA POWER SYSTEM OPERATION 147

REFERENCES [24] A. G. Bakirtzis and P. N. Biskas, “A decentralized solution to the DC-OPF

of interconnected power systems,” IEEE Trans. Power Syst., vol. 18, no. 3,
[1] A. J. Wood, Power Generation, Operation, and Control. 2nd ed. New pp. 1007–1013, Aug. 2003.
York, NY, USA: Wiley, 1996. [25] A. Kargarian, Y. Fu, and Z. Li, “Distributed security-constrained unit
[2] Joint operating agreement among and between New York Indepen- commitment for large-scale power systems,” IEEE Trans. Power Syst.,
dent System Operator INC., (“NYISO”) and PJM interconnection, vol. 30, no. 4, pp. 1925–1936, Jul. 2014.
L.L.C. (“PJM”). May 2007. [Online]. Available: https://www.pjm.com/∼ [26] A. Ahmadi-Khatir, A. J. Conejo, and R. Cherkaoui, “Multi-area unit
/media/documents/agreements/nyiso-pjm.ashx scheduling and reserve allocation under wind power uncertainty,” IEEE
[3] Joint operating agreement between the Midwest Independent Trans- Trans. Power Syst., vol. 29, no. 4, pp. 1701–1710, Jul. 2014.
mission System Operator, INC., and PJM Interconnection, L.L.C. Dec. [27] M. Doostizadeh, F. Aminifar, H. Lesani, and H. Ghasemi, “Multi-area
2003. [Online]. Available: http://www.pjm.com/∼/media/documents/ market clearing in wind-integrated interconnected power systems: A
agreements/joa-complete.ashx fast parallel decentralized method,” Energy Convers. Manage., vol. 113,
[4] M. Shahidehpour, H. Yamin, and Z. Li, Market Operations in Electric pp. 131–142, Apr. 2016.
Power Systems: Forecasting, Scheduling, and Risk Management. New [28] Y. Fu, C. Wang, W. Tian, and M. Shahidehpour, “Integration of large-
York, NY, USA: Wiley-Interscience, 2002. scale offshore wind energy via VSC-HVDC in day-ahead scheduling,”
[5] G. Cohen, “Auxiliary problem principle and decomposition of optimiza- IEEE Trans. Sustain. Energy, vol. 7, no. 2, pp. 535–545, Apr. 2016.
tion problems,” J. Optim. Theory Appl., vol. 32, no. 3, pp. 277–305, [29] Y. Fu, Z. Li, and L. Wu, “Modeling and solution of the large-scale security-
Nov. 1980. constrained unit commitment,” IEEE Trans. Power Syst., vol. 28, no. 4,
[6] J. Kabouris and F. D. Kanellos, “Impacts of large-scale wind penetration on pp. 3524–3533, Nov. 2013.
designing and operation of electric power systems,” IEEE Trans. Sustain. [30] J. Wang, M. Shahidehpour, and Z. Li, “Security-constrained unit com-
Energy, vol. 1, no. 2, pp. 107–114, Jul. 2010. mitment with volatile wind power generation,” IEEE Trans. Power Syst.,
[7] Y. Wang, S. Zhao, Z. Zhou, A. Botterud, Y. Xu, and R. Chen, “Risk vol. 23, no. 3, pp. 1319–1327, 2008.
adjustable day-ahead unit commitment with wind power based on chance [31] L. Baringo and A. J. Conejo, “Correlated wind-power production and
constrained goal programming,” IEEE Trans. Sustain. Energy, vol. 8, no. 2, electric load scenarios for investment decisions,” Appl. Energy, vol. 101,
pp. 530–541, Apr. 2017. pp. 475–482, 2013.
[8] Z. Wu, P. Zeng, X. P. Zhang, and Q. Zhou, “A solution to the chance- [32] M. Yesilbudak, “Clustering analysis of multidimensional wind speed data
constrained two-stage stochastic program for unit commitment with using k-means approach,” in Proc. IEEE Int. Conf. Renew. Energy Res.
wind energy integration,” IEEE Trans. Power Syst., vol. 31, no. 6, Appl., Birmingham, UK, Nov. 2016, pp. 961–965.
pp. 4185–4196, Nov. 2016.
[9] D. Bertsimas, E. Litvinov, X. A. Sun, J. Zhao, and T. Zheng, “Adaptive ro-
bust optimization for the security constrained unit commitment problem,”
IEEE Trans. Power Syst., vol. 28, no. 1, pp. 52–63, Feb. 2013.
[10] R. Jiang, J. Wang, and Y. Guan, “Robust unit commitment with wind Mojtaba Khanabadi (S’12) received the B.S. degree
power and pumped storage hydro,” IEEE Trans. Power Syst., vol. 27, in electrical engineering from Imam Khomeini Inter-
no. 2, pp, 800–810, May 2012. national University, Qazvin, Iran, in 2009 and the
[11] A. Tuohy, P. Meibom, E. Denny, and M. O’Malley, “Unit commitment M.S. degree in electrical engineering from Univer-
for systems with significant wind penetration,” IEEE Trans. Power Syst., sity of Tehran, Tehran, Iran, in 2012. He is currently
vol. 24, no. 2, pp. 592–601, May 2009. working toward the Ph.D. degree in electrical en-
[12] A. Papavasiliou, S. S. Oren, and R. P. O’Neill, “Reserve requirements gineering at Mississippi State University, Starkville,
for wind power integration: A scenario-based stochastic programming MS, USA. His research interests include power sys-
framework,” IEEE Trans. Power Syst., vol. 26, no. 4, pp. 2197–2206, tems optimization and economics, congestion man-
Nov. 2011. agement, and energy management systems.
[13] F. Bouffard and F. D. Galiana, “Stochastic security for operations planning
with significant wind power generation,” IEEE Trans. Power Syst., vol. 23,
no. 2, pp. 306–316, May 2008.
[14] C. Uckun, A. Botterud, and J. R. Birge, “An improved stochastic unit
commitment formulation to accommodate wind uncertainty,” IEEE Trans. Yong Fu (M’05–SM’13) received the B.S. and
Power Syst., vol. 31, no. 4, pp. 2507–2517, Jul. 2016. M.S. degrees in electrical engineering from
[15] L. Wu, M. Shahidehpour, and T. Li, “Stochastic security-constrained unit Shanghai Jiaotong University, Shanghai, China, in
commitment,” IEEE Trans. Power Syst., vol. 22, no. 2, pp. 800–811, 1997 and 2002, respectively, and the Ph.D. degree in
May 2007. electrical engineering from Illinois Institute of Tech-
[16] C. Wang and Y. Fu, “Fully parallel stochastic security-constrained unit nology, Chicago, IL, USA, in 2006. He is currently
commitment,” IEEE Trans. Power Syst., vol. 31, no. 5, pp. 3561–3571, an Associate Professor in the Department of Electri-
Sep. 2016. cal and Computer Engineering at Mississippi State
[17] J. Macqueen, “Some methods for classification and analysis of multivariate University. His research interests include power sys-
observations,” in Proc. 5th Berkeley Symp. Math. Stat. Probab., University tem optimization and economics, renewable energy
of California Press, 1967, pp. 281–297. integration, and smart grid. He has been awarded the
[18] Y. Wang, Y. Lin, and D. Kirschen, “Scenario reduction with submodular Tennessee Valley Authority Endowed Professorship in Power Systems Engi-
optimization,” IEEE Trans. Power Syst., vol. 32, no. 3, pp. 2479–2480, neering. He has received the NSF CAREER Award in 2012.
May 2017.
[19] J. Dupacová, N. Gröwe-Kuska, and W. Römisch, “Scenario reduction in
stochastic programming: An approach using probability metrics,” Math.
Program, vol. 95, no. 3, pp. 493–511, Mar. 2003.
Lin Gong (S’12) received the B.S. degree in elec-
[20] N. Gröwe-Kuska, H. Heitsch, and W. Römisch, “Scenario reduction and
trical engineering from Wuhan University, Wuhan,
scenario tree construction for power management problems,” in Proc.
China, in 2005 and the M.S. degree in electrical
IEEE Bologna Power Tech, Bologna, Italy, pp. 1–7, Jun. 2003.
engineering from Illinois Institute of Technology,
[21] W. R¨omisch and H. Heitsch, “Scenario reduction algorithms in stochastic
Chicago, IL, USA, in 2014. He is currently working
programming,” Comput. Optim. Appl., vol. 24, no. 2/3,pp. 187–206, 2003.
toward the Ph.D. degree in electrical engineering at
[22] B. H. Kim and R. Baldick, “Coarse-grained distributed optimal power Mississippi State University, Starkville, MS, USA.
flow,” IEEE Trans. Power Syst., vol. 12, no. 2, pp. 932–939, May 1997.
His research interests include power systems opti-
[23] R. Baldick, B. H. Kim, C. Chase, and Y. Luo, “A fast distributed imple-
mization and economics, and parallel computing al-
mentation of optimal power flow,” IEEE Trans. Power Syst., vol. 14, no. 3,
gorithms and applications.
pp. 858–864, Aug. 1999.